Seonglae ChoKeys go to the hippocampus for quick access. Values go tot the Neocortex for high-fidelity storage.
The hippocampus specializes in storing and retrieving keys (addresses), while the neocortex specializes in storing values (content), forming complementary learning systems. Keys are optimized for search discriminability, while values are optimized for storage fidelity in separate learning spaces
Teyler & DiScenna (1986)
The hypothesis that the hippocampus functions as an index pointing to memory contents stored in the neocortex
A proof of concept connecting machine learning's key-value mechanism with existing psychological and neurological theories (hippocampal indexing, complementary learning systems, etc.). It builds key-value association matrices through Hebbian theory and retrieves value vectors through attention mechanisms (similarity kernels + softmax, etc.) using weighted summation.